Masked diffusion enables coherent beat tracking

2026-08-05

AI Summary

Existing beat tracking methods often produce consecutive downbeats or unstable outputs due to their inability to model multiple plausible metrical structures. This work proposes an improved masked diffusion model that effectively explores diverse beat grids during iterative inference by introducing independent masking mechanisms for beats and downbeats, a dynamically balanced mask scheduling strategy, and a stride-aware peak selection method. The proposed approach significantly enhances the consistency and accuracy of beat prediction, outperforming current state-of-the-art methods on standard benchmarks and effectively mitigating anomalous beat-tracking behaviors.

Abstract

Current neural networks for beat tracking generate invalid outputs, such as consecutive downbeats and erratic tempo changes, even when these are not present in the training data. Heavy post-processing techniques can alleviate these problems, but the original cause of this inconsistent behaviour remains unknown. We hypothesise that it stems from inadequate modelling of multiple plausible output beat grids, resulting in an invalid mixture of competing interpretations. We propose a masked diffusion approach that properly models multiple outputs and enables the model to build coherent predictions through iterative inference. We devise three modifications to standard masked diffusion that enable its application to beat tracking: independent masking of beats and downbeats during training and inference, a balanced masking scheduler for inference, and peak-picking across inference steps. Our approach reduces erratic behaviours and improves beat-tracking performance.

Problem

• beat tracking
• neural networks
• erratic tempo changes
• coherent predictions
• invalid outputs

Innovation

• masked diffusion
• beat tracking
• coherent prediction
• multiple output modeling
• tempo consistency